Quick answer 54% of the 6,004 open roles state an hourly rate. A further 14% give only an "up to" ceiling or a per-task rate, and 32% state no pay at all. Updated 3 November 2026
Por plataforma
The platforms with at least 20 open roles, ordered by the share that state an hourly rate. It runs from 100% at 10x Team to 0% at CNTXT AI. 4 platforms state a rate on every role. 5 state no pay on any role.
Por área
Roles grouped by their main kind of work. Law roles are the most likely to state a rate (73%), and Data annotation roles the least (25%). Pay data has what the rates are.
Por país
The countries named by the most roles. A role open to several countries counts under each. Roles open worldwide are shown on their own.
Over time
How many roles say what they pay
Share of open roles by how the posting states pay
On 3 Nov, the largest groups were States an hourly rate (53.6%), Only an "up to" or per-task rate (14.2%), States no pay (32.2%).
Each open role is counted once, by what its posting says about pay. Recorded daily from 21 October 2026.
Mostrar os números
| Quando | States an hourly rate | Only an "up to" or per-task rate | States no pay |
|---|---|---|---|
| 21 out | 53.4% | 15.5% | 31.1% |
| 22 out | 52% | 14.4% | 33.5% |
| 23 out | 54% | 14.2% | 31.8% |
| 24 out | 53.9% | 14.4% | 31.7% |
| 25 out | 56.2% | 14% | 29.8% |
| 26 out | 55.6% | 14.1% | 30.3% |
| 27 out | 55.4% | 14.3% | 30.3% |
| 28 out | 55.4% | 14.4% | 30.1% |
| 29 out | 55.8% | 14.5% | 29.7% |
| 30 out | 55.7% | 14.5% | 29.7% |
| 31 out | 55.9% | 14.6% | 29.6% |
| 1 nov | 53.8% | 13.9% | 32.3% |
| 2 nov | 53.7% | 14.2% | 32.1% |
| 3 nov | 53.6% | 14.2% | 32.2% |
How we count
- Roles counted: the open AI training roles on the platforms' own job boards at the latest hourly update. Paid research studies and talent pools are left out.
- Each role once: a role is in exactly one group. States an hourly rate: the posting gives an hourly amount or range. Only an "up to" or per-task rate: it gives a ceiling ("up to $50 an hour"), or pay per task or per audio hour. States no pay: no figure at all.
- Not the same as the pay figures: pay data counts each distinct posting once, so roles repeated per country or language count once there and every time here. The two pages will therefore show different totals.
- Limits: we see what platforms publish. A platform may tell you the rate after you apply, or set it per project. That is not the same as paying badly.
Download the data
Free to use with credit under CC BY 4.0. Cite as: aitrainers.jobs (2026). Pay transparency in AI training jobs. https://aitrainers.jobs/pay-transparency